Tuesday, March 07, 2006

Typology and Classification - what are they good for?

There is occasional discussion among my friends and colleagues about classifying rock piles. Mavor and Dix have a simple classification of types of piles they encountered in the Falmouth area. The Gage's have some classifications on their website, and I have created a couple of classificatons. One is online (with some missing photos) [here]

Anyway, I have argued with people who create arbitrary categories (abstract categories for which there are no known examples) about whether these should be eliminated from the classification until examples are found. And it is easy to imagine getting into an extended argument like: "no that's a pile of rocks on a support boulder but with damage, so the rocks are falling off and trailing to the ground...and...." where there is really no basis for argument or for conclusion.

When you have a classification scheme, here are some questions worth trying to answer:
  • What good is the classification?
  • How would you verify it as a legitimate classification?
  • How did you come up with that particular classification rather than another?
For it to be science (according to -was it- Putnam, the philosopher?) it has to be verifiable. So these are non-trivial questions. What is important to avoid is an empty discussion without addressing such fundamental questions up front.

Since I am guilty of having created a classification (or two) let me take a stab at question #1, what good is the classification? Here is an answer I find satisfying: with a classification one can go the next step and create a distributional plot (like the graphic included above). This plot shows which types of various rock piles are found in which towns nearby. With such a plot it is possible to argue that different types were created in different places at different times by different people. Also it is possible to group different types and say they were probably created by the same people in the same general area, perhaps within the same time period. It is also possible to validate the classification - at least to the extent that you have enough data, because if the distributional plot has no logic then that argues against using the classification. To my eyes the classification illustrated above has some validity. I am not going to try to quantify it.

So here is my challenge to others who propose classifications: do a distributional plot and show me that the classification has validity. After that we can see what can be deduced. Without this sort of follow-through it is just a matter of opinion. To give one last example for clarity: suppose a type is proposed that is found everywhere in the study area. From that, almost nothing can be concluded. So it is in the corellation of type with other attributes (I am focusing on location) that these things have meaning.

No comments :